Group Members : Tharun Prabhakar, Niranjan Kumar, Sumit Kumar¶

From above charts we can see the data divided among the regions of USA and south is the region with most providers, while west has the least number of providers.

• Above we see the shape of the dataframe and see that df has 163K clean rows to be analysed, and the categorical variables in the df shows how many different variation of data exist in df.
• This diversity underscores the complexity of the dataset and necessitates a careful examination of these categorical variables for a comprehensive analysis.
• Hence below we are going to calculate z score for 20 combination of Categorical and Continous variable, these 20 combination will give us the 20 feature that we are seeking from the data.

Summary: These feature represents the standardized score of the total number of discharges, Average Covered Charges, Average Total Payments, Average Medicare Payments by a provider region, DRG Definition, Hospital Referral Region, Provider City, Provider Name respectively.

Insight:

1. An unusually high or low number of discharges, could indicate irregularities in patient admissions or discharge practices. This could be a sign of fraudulent activities like patient churning or unnecessary admissions.

2. Extreme values in this features may indicate that a supplier charges much more or less than competitors in the same location. This might suggest service overcharging, service undercharging as part of a bribe fraudulent activity, or service fraud.

3. Total payments that are unusually high or low may indicate billing problems. High levels may indicate overutilization of services, while low values may indicate a lack or unreported services.

4. Significant differences in Medicare payments can be indicators of fraudulent activity, such as false claims or services not provided. Both abnormally high and unusually low results require additional investigation.

Discharges Total_z-value_PR: A high positive z-value shows that a provider has more discharges than the region's standard, which might imply significant patient turnover or over-admissions in comparison to competitors. A negative z-value indicates fewer discharges than predicted, which might be related to lower admission rates or stricter admission requirements.

Covered Charges Average_z-value_PR: Positive z-values show that a provider's charges are greater than the regional average,  which could indicate overcharging or higher service costs. Negative numbers may indicate undercharging as a result of reduced operating expenses or alternative pricing methods.

Average Total Payments_z-value_PR: This measure compares the provider's total payments to the regional average. A greater z-value may indicate possible overutilization or overbilling of services, whilst a lower value may imply underbilling or more efficient service supply.

Medicare Average Payments_z-value_PR: Positive z-values indicate that the provider receives more from Medicare than the area average, which might be attributable to increased service utilisation or invoicing for more sophisticated treatments. Negative z-values may imply decreased reliance on Medicare, either as a result of a patient mix with less Medicare coverage or more cautious billing procedures.

Total Discharges_z-value_DRG: A high z-value, like the 2.73 for the first entry, suggests significantly higher patient discharges compared to others in the same DRG category, which might indicate high treatment frequency or potential overuse of services. Negative values, like -0.78, indicate lower than average discharges, which could mean more selective patient treatment or potentially insufficient care.

Average Covered Charges_z-value_DRG: Positive z-scores imply that a provider's charges are higher than the average for that DRG category, which could suggest overpricing or specialization in more complex procedures. Negative scores suggest lower-than-average charges, which could reflect more efficient care, lower cost structures, or potentially undercharging.

Average Total Payments_z-value_DRG: This metric reflects whether the provider's total payments are high or low relative to the DRG average. For example, a z-score of -1.03 indicates payments are lower than average, which may raise questions about the provider's service quality or billing practices.

Average Medicare Payments_z-value_DRG: These z-scores indicate how a provider's Medicare payments compare to the average for the DRG. Negative values, such as -1.15, could suggest a provider is receiving less Medicare reimbursement than peers, possibly due to lower service utilization or conservative billing.

Total Discharges_z-value_Reff: A high z-score, like the 1.36 for the first entry, indicates that a hospital has a higher number of discharges compared to the average in its referral region. This could suggest a higher patient volume or a tendency to admit and discharge more frequently than other hospitals in the area.

Average Covered Charges_z-value_Reff: This measures how a hospital's average charges for services compare to the regional average. A positive z-score indicates higher than average charges, which might reflect a hospital's pricing strategy or suggest that it provides more specialized or intensive services.

Average Total Payments_z-value_Reff: A negative z-score here suggests that a hospital is receiving lower total payments than the regional average, possibly indicating efficiency in service delivery or challenges in obtaining reimbursements.

Average Medicare Payments_z-value_Reff: Negative z-scores suggest that a hospital's Medicare payments are less than the regional average. This could raise questions about the hospital's Medicare patient demographics, the mix of services provided to Medicare patients, or its Medicare billing practices.

Total Discharges_z-value_ProvCity: Positive z-scores, like 0.83, suggest that a provider has more discharges than the average for their city. This might indicate higher patient throughput. Negative z-scores, such as -0.82, show fewer discharges than the city average, potentially pointing to a smaller operational scale or more selective patient admission.

Average Covered Charges_z-value_ProvCity: The z-scores indicate whether a provider's charges are above or below the city average. Most values shown are negative, suggesting these providers generally charge less than the city average for their services.

Average Total Payments_z-value_ProvCity: These z-scores compare the total payments received to the city average. Most providers in the table have negative z-scores, implying they receive less in payments compared to the city average, which could be due to a variety of factors including the types of services provided, patient demographics, or efficiency in service delivery.

Average Medicare Payments_z-value_ProvCity: Similar to total payments, these z-scores compare Medicare payments against the city average. Negative values, such as -0.54, suggest a provider's Medicare reimbursements are below the city average, possibly indicating less reliance on Medicare billing, fewer services provided to Medicare patients, or more conservative billing practices.

Total Discharges_z-value_ProvName: Positive z-scores (like 0.60 for the first entry) suggest a provider has a higher number of discharges compared to the average for that provider name. This might suggest a higher patient volume or different operational practices. Negative z-scores indicate fewer discharges, which could be due to various factors such as smaller facility size or patient demographics.

Average Covered Charges_z-value_ProvName: This metric reflects the provider's pricing relative to the average for that provider name. Most values in the table are negative, which suggests these particular providers charge less for services than the average for their provider name group.

Average Total Payments_z-value_ProvName: Negative z-scores across the board suggest that these providers receive lower total payments than the average for their provider name group. This could indicate more efficient service provision, a different payer mix, or possibly an underutilization of services.

Average Medicare Payments_z-value_ProvName: Similar to total payments, these z-scores reflect how Medicare payments to the provider compare to the average for their provider name group. Negative scores, such as -0.53 for the third entry, could suggest lower reimbursement rates from Medicare, potentially due to a variety of billing practices or service offerings.

EXPLAIN

• PCA:
  • PCA is used for dimensionality reduction in high dimentional datasets.
  • PCA can be useful for highlighting outliers.
• KNN:
  • K-Nearest Neighbors (KNN) is a versatile algorithm in machine learning, used both in supervised and unsupervised learning contexts.
  • In PyOD, KNN is used for anomaly detection without relying on target labels. Instead, it calculates the distance of each data point to its k nearest neighbors and uses these distances to define outlier scores.

Results¶

PCA and KNN Outlier Detection by Region:¶

Midwest:

Non-Outliers (0.00): Lower average values for covered charges, total payments, and Medicare payments, with a mean of 41.07 total dischgrses. Outliers (1.00): Significantly higher averages in all financial categories and a mean of 98.91 total discharges.

Northeast:¶

Non-Outliers: Moderate average values for covered charges, total payments, and Medicare payments, with a mean of 43.04 total discharges. Outliers: Highest average values among all regions for outliers, with a substantial mean of 96.80 total discharges.

South:¶

Non-Outliers: Similar to the Midwest in terms of financial figures, with a mean of 40.96 total discharges. Outliers: High averages, but lower than the Northeast outliers, with a mean of 99.94 total discharges.

West:¶

Non-Outliers: The lowest average values for covered charges and total payments, with a mean of 32.45 total discharges. Outliers: Very high average values, particularly for covered charges, with a mean of 79.30 total discharges.

Midwest:¶

Non-Outliers: Similar pattern as in PCA, with moderate average values and a mean of 40.88 total discharges. Outliers: Extremely high averages in all financial categories and a mean of 117.70 total discharges.

Northeast:¶

Non-Outliers: Higher average values than the Midwest non-outliers and a mean of 41.54 total discharges. Outliers: Very high averages, especially for covered charges, with a mean of 108.31 total discharges.

South:¶

Non-Outliers: Lower average values compared to the Northeast non-outliers, with a mean of 40.54 total discharges. Outliers: High averages, slightly less than the Northeast outliers, with a mean of 118.08 total discharges.

West:¶

Non-Outliers: The lowest average values for covered charges and total payments, similar to PCA, with a mean of 31.65 total discharges. Outliers: Extremely high averages, with the highest average total payments and a mean of 79.99 total discharges.

Interpretation:¶

Outliers vs Non-Outliers: For both PCA and KNN, the outliers have significantly higher average charges and payments compared to non-outliers. The mean total discharges for outliers are also higher, suggesting that outliers may be associated with higher volume facilities or more complex cases that justify higher chargo PCA.

Comparison between PCA and KNN:¶

While both methods identify outliers with higher charges and payments, there is a noticeable difference in the magnitude. Outliers detected by PCA have even higher average charges and payments than those detected by KNN, which could suggest that PCA is more sensitive to extreme values or that it captures different aspects of the data that lead to a classification as an outlier. The average number of discharges for outliers is also higher in the KNN method compared to PCA. There is consistency in the pattern of outliers having higher values across both methods, but the KNN method seems to flag more extreme values as outliers compared to PCA, particularly in the Midwest and West regions.

Business Implications:¶

The marked difference between outliers and non-outliers in terms of financial figures could point to potential cases of fraud, waste, or abuse, or they could be legitimate cases where higher costs are justified. The differences between PCA and KNN in identifying outliers could suggest that combining these methods may provide a more comprehensive approach to detecting potential fraud. The higher total discharges among outliers may warrant further investigation to determine if this is due to legitimate reasons such as a specialization in certain high-cost treatments or if it's an indication of potential overcharging or unnecessary services being render The Northeast region appears to have the highest average values for outliers detected by both methods, suggesting this region may require more focused audit and fraud detection efforts. Outliers identified by KNN in the Midwest show particularly high average covered charges, which could warrant further investigation. The consistency of the West region having the lowest non-outlier values across both methods might indicate a different pricing structure or market conditions in that region. The differences in outlier detection between PCA and KNN underscore the importance of using multiple methods for a more robust fraud detection strategy.ed. In summary, these results suggest that there are distinct differences between the data points flagged as outliers and those that are not, according to both PCA and KNN algorithms. Further investigation into the reasons behind these discrepancies would be necessary to determine if they are due to fraud or other factors.